ACXNet hybrid deep learning model for cross task mental workload estimation using EEG neural manifolds

Abinaya, G.; Dinakaran, K. · 2025 · Crossref

DOI: 10.1038/s41598-025-19144-x

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of accurate, cross-task mental workload (MWL) estimation using electroencephalogram (EEG) signals, a critical need for preventing cognitive overload in high-stakes environments like aviation and healthcare. While traditional methods rely on subjective self-reports or task-specific performance metrics, and existing deep learning models often struggle with generalizability across different tasks without subject-specific calibration, this study proposes a novel hybrid deep learning model called ACXNet. The primary motivation is to develop a robust system that can estimate MWL levels without prior subject-specific calibration or task-specific pre-labeled training data, thereby enhancing the scalability and real-world applicability of brain-computer interfaces. The methodology involves a three-stage pipeline integrating an autoencoder, a Convolutional Neural Network (CNN), and XGBoost. The model was validated using the Simultaneous Task EEG Workload (STEW) dataset, which contains recordings from 48 participants performing the SIMKAP multitasking protocol and a resting condition. EEG data was collected using an Emotiv EPOC device at 128 Hz with 14 channels. Preprocessing included artifact removal via Independent Component Analysis (ICA) and segmentation into 2-second windows (256 samples). The ACXNet architecture first employs an autoencoder for unsupervised dimensionality reduction, compressing the high-dimensional input (3584 features) into a 128-feature latent space to preserve critical information while reducing computational complexity. This compressed representation is then processed by a CNN to capture spatial-temporal dependencies, followed by an XGBoost classifier for final binary classification of low versus high mental workload. The results demonstrate that ACXNet outperforms existing methods in cross-task estimation. Specifically, the model achieved an average accuracy of 92.10% for the SIMKAP task and 89.94% for the "No task" (resting) condition. These findings indicate that the hybrid approach effectively leverages the autoencoder for feature extraction, the CNN for pattern recognition, and XGBoost for robust classification, significantly improving the precision and robustness of MWL estimation compared to conventional techniques. The significance of this work lies in its ability to provide a scalable solution for intelligent human-computer interaction systems. By eliminating the need for extensive task-specific calibration, ACXNet opens new avenues for deploying real-time cognitive monitoring in healthcare, aviation, and other domains where sustained attention is required. The study contributes to the field by demonstrating that combining unsupervised feature learning with gradient boosting can enhance both the accuracy and interpretability of EEG-based cognitive state assessment, addressing key limitations in current cross-task generalization research.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data